iMatch 2020.3.6

[ben]

Thanks Mario, for the long list of improvements!!
2020.3.6

[thrinn]

Downloaded and installed without problems.
Impressive list of improvements!

[mastodon]

Handling of face tags is very much improved. Thank you!

[Aubrey]

Long list of improvements... these will be most useful.

Thanks,
Aubrey.

[Mario]

Thanks. The past two weeks were hell and I'm walking on my gums right now 
And there is of course all the Corona unpleasantness.

I hope to reduce daily hours for a week or two now.

[Aubrey]

Looks like you've resolved the "Face Annotation: buttons don't react to clicks" for me. Everything is working (I didn't comment on problem, as I'd no idea workflow that was causing it!)

Maybe it's subjective, but face annotations appears to be picking up known faces even better.

Thanks, and take a break. Most importantly you and family keep well.

Aubrey.

[Mario]

Looks like you've resolved the "Face Annotation: buttons don't react to clicks" for me. Everything is working (I didn't comment on problem, as I'd no idea workflow that was causing it!)

Maybe it's subjective, but face annotations appears to be picking up known faces even better.

Thanks for the feedback. Since I could never reproduce the "click not working sometimes" reported by Mees and two (?) other users, I was not sure.
My guess was that this was a focus issue. But I'm still not sure. Hard to fix something that does not happen on your PC.

As written in the release notes, I have improved the face recognition algorithm in this release, the part that weights the age difference between trained faces and analyzed faces.
The new ability to review and control which faces IMatch uses to identify people will help improve results, too.
Especially for users who have images of persons at a wide variety of ages, especially childhood. While the IMatch AI algorithms are relatively stable and age-insensitive, training more faces for children can help. Because children face geometries change quite a lot in the first 10 or so years. Not everything can be 'built-into' an algorithm.

Commercial face recognition research is mostly based on adults - because this is where companies like Facebook, Google and others make their revenue. From tracking us (adults) all over the web and in the real world by our faces in images we or others upload... 
Children are (commercially) not that interesting. And police, military and intelligence face recognition research also concentrates on adults, obviously.

[mastodon]

The review of trained faces is very good. I have found some misstrained faces, too, that are totally false (not that person). I can remove them from these from trained faces, but they still have false face tag. How can I find these pictures, to correct these face tags?

[Mario]

Untraining a face has no impact on already assigned faces.
You can make IMatch 'recluster' the person, to find matching faces via the corresponding command in the People View.
See Finding Faces for Persons
This will then use the new set of trained faces to find matching faces for the person. It will not change confirmed faces, it only looks at unconfirmed and faces and faces not assigned to any person.

[abgestumpft]

Hi Mario,

I just tested a little bit:
1. Face Recognition: works really better now (better hitrate). The new options like reject& delete faces speed up a lot the process of assigning correct faces. Also reviewing/correcting trained faces is a great feature. The whole process of Face assignment is now much faster... Thanks! 

2. Importing images that have unnamed faces now all end up in the "Unkown Persons" "Person" -> great


What I found is not working:
a) When using WIC Codecs and importing Olmypus RAW in portrait orientation, they are still shown in landscape orientation after import -> So I keep using the workaround to use iMatch's internal libraw which is working fine

b) One minor thing: If a person is junger than 1 month in a picture it shows age: () -> this should show the age in days


Another thing I found out yesterday (with 2020.3.4 version):
When a person is not correctly recognized it was often the case that the face box was very close under the lips. By moving the box a little bit to inculde (parts) of the chin often resulted in the correct face detected.
This has obviously also been fixed by 2020.3.6 -> when I delete such a face box (created with 2020.3.4) and rescan, the new box includes more parts of the chin which results in better hitrate.
This means if somebody has still unassigned / unconfirmed faces it should make sense to remove them and have the faces re-detected with Version 2020.3.6.

[Mario]

Please add one bug report for each bug found in the bug report board.
Bugs mentioned in passing in a 'container thread' like this will be forgotten otherwise.
Having a separate thread also makes tracking easier and I can link to the bug report when I document the bug fix in the release notes.

I'm not sure about ORF files anymore. What do they do with these files? Why does Olympus not provide a WIC codec to their customers?
Maybe your files are the exception from the exception and neither the "apply rotation as indicated in EXIF" or the "ignore EXIF orientation work". Which WIC codecs do you use? FPC? Standard Windows? Latest from the shop (separate free download).

Google for Get Raw Image Extension - Microsoft Store

These WIC codecs are based on LibRaw and hence should work with 2020.3.6. If not, I have no further idea. I have already implemented all exceptions for ORF files.

Send me a sample file to support email address (with a link back to your bug report or this thread).

[GrantRobertson]

This does, indeed fix my issue with Ctrl-Shift-T not working, reaffirming our suspicions that it was caused by the File Window loosing focus issue. Now, when I use that hotkey, the file stays selected and everything works as expected.